Abstract: This paper designs and realizes the adaptive shot segmentation algorithm. The shot change types and the factors that affect the accuracy are classified. And the complicated situation is considered as a classification problem. Every level classifier may choice suitable algorithms to distinguish one class from the other. Then they are organized into a tree-like classifier and make the performance of system better. The results of experiments on four advertisements prove that this new adaptive algorithm can acclimatize itself to various situations to detect both abrupt shot changes and gradual shot changes and reduce effectively false rate on the basis of ensuring hit rate.

Key words: Video retrieval, Shot change, Abrupt change, Gradual change, Shot segmentation

摘要： 提出并实现了一种自适应的镜头分割系统。将镜头切换类型和影响镜头分割准确性的因素进行分类，把要处理的复杂情况看作是多类别的分类问题。每一级分类器处理的都是两类问题，根据分类的不同目的选择合适的算法将两类区分开。将这些分类器以树形结构结合起来，使系统的性能达到最优。选择了4个有代表性的广告视频片段进行实验。结果表明，自适应镜头分割算法可以适应不同的情况，同时检测突变切换和渐变切换，在保证检全率的基础上有效降低了误检率。

关键词: 视频检索, 镜头切换, 突变切换, 渐变切换, 镜头分割